arXiv subjects

Dong Li

Publications and source records attributed to Dong Li.

At least 145 records · Page 8Linked to original sources

arXiv2025

Machine Learning-Guided Memory Optimization for DLRM Inference on Tiered Memory

Deep learning recommendation models (DLRMs) are widely used in industry, and their memory capacity requirements reach the terabyte scale. Tiered memory architectures provide a cost-effective solution but introduce challenges in embedding-vector placement due to complex embedding-access patterns. We propose RecMG, a machine learning (ML)-guided system for vector caching and prefetching on tiered memory. RecMG accurately predicts accesses to embedding vectors with long reuse distances or few reuses. The design of RecMG focuses on making ML feasible in the context of DLRM inference by addressing unique challenges in data labeling and navigating the search space for embedding-vector placement. By employing separate ML models for caching and prefetching, plus a novel differentiable loss function, RecMG narrows the prefetching search space and minimizes on-demand fetches. Compared to state-of-the-art temporal, spatial, and ML-based prefetchers, RecMG reduces on-demand fetches by 2.2x, 2.8x, and 1.5x, respectively. In industrial-scale DLRM inference scenarios, RecMG effectively reduces end-to-end DLRM inference time by up to 43%.

cs.PF↗

arXiv2025

Adaptive Phase Shift Information Compression for IRS Systems: A Prompt Conditioned Variable Rate Framework

Intelligent reflecting surfaces (IRSs) have become a vital technology for improving the spectrum and energy efficiency of forthcoming wireless networks. Nevertheless, practical implementation is obstructed by the excessive overhead associated with the frequent transmission of phase shift information (PSI) over bandwidth-constrained control lines. Current deep learning-based compression methods mitigate this problem but are constrained by elevated decoder complexity, inadequate flexibility to dynamic channels, and static compression ratios. This research presents a prompt-conditioned PSI compression system that integrates prompt learning inspired by large models into the PSI compression process to address these difficulties. A hybrid prompt technique that integrates soft prompt concatenation with feature-wise linear modulation (FiLM) facilitates adaptive encoding across diverse signal-to-noise ratios (SNRs), fading kinds, and compression ratios. Furthermore, a variable rate technique incorporates the compression ratio into the prompt embeddings through latent masking, enabling a singular model to adeptly balance reconstruction accuracy. Additionally, a lightweight depthwise convolutional gating (DWCG) decoder facilitates precise feature reconstruction with minimal complexity. Comprehensive simulations indicate that the proposed framework significantly reduces NMSE compared to traditional autoencoder baselines, while ensuring robustness across various channel circumstances and accommodating variable compression ratios within a single model. These findings underscore the framework's promise as a scalable and efficient solution for real-time IRS control in next-generation wireless networks.

eess.SP↗

arXiv2025

Fairness-Aware Computation Offloading in Wireless-Powered MEC Systems with Cooperative Energy Recycling

In this paper, cooperative energy recycling (CER) is investigated in wireless-powered mobile edge computing systems. Unlike conventional architectures that rely solely on a dedicated power source, wireless sensors are additionally enabled to recycle energy from peer transmissions. To evaluate system performance, a joint computation optimization problem is formulated that integrates local computing and computation offloading, under an alpha-fairness objective that balances total computable data and user fairness while satisfying energy, latency, and task size constraints. Due to the inherent non-convexity introduced by coupled resource variables and fairness regularization, a variable-substitution technique is employed to transform the problem into a convex structure, which is then efficiently solved using Lagrangian duality and alternating optimization. To characterize the fairness-efficiency tradeoff, closed-form solutions are derived for three representative regimes: zero fairness, common fairness, and max-min fairness, each offering distinct system-level insights. Numerical results validate the effectiveness of the proposed CER-enabled framework, demonstrating significant gains in throughput and adaptability over benchmark schemes. The tunable alpha fairness mechanism provides flexible control over performance-fairness trade-offs across diverse scenarios.

cs.IT↗

arXiv2025

Logic-informed reinforcement learning for cross-domain optimization of large-scale cyber-physical systems

Cyber-physical systems (CPS) require the joint optimization of discrete cyber actions and continuous physical parameters under stringent safety logic constraints. However, existing hierarchical approaches often compromise global optimality, whereas reinforcement learning (RL) in hybrid action spaces often relies on brittle reward penalties, masking, or shielding and struggles to guarantee constraint satisfaction. We present logic-informed reinforcement learning (LIRL), which equips standard policy-gradient algorithms with projection that maps a low-dimensional latent action onto the admissible hybrid manifold defined on-the-fly by first-order logic. This guarantees feasibility of every exploratory step without penalty tuning. Experimental evaluations have been conducted across multiple scenarios, including industrial manufacturing, electric vehicle charging stations, and traffic signal control, in all of which the proposed method outperforms existing hierarchical optimization approaches. Taking a robotic reducer assembly system in industrial manufacturing as an example, LIRL achieves a 36.47\% to 44.33\% reduction at most in the combined makespan-energy objective compared to conventional industrial hierarchical scheduling methods. Meanwhile, it consistently maintains zero constraint violations and significantly surpasses state-of-the-art hybrid-action reinforcement learning baselines. Thanks to its declarative logic-based constraint formulation, the framework can be seamlessly transferred to other domains such as smart transportation and smart grid, thereby paving the way for safe and real-time optimization in large-scale CPS.

cs.LG↗

arXiv2025

AMD-Hummingbird: Towards an Efficient Text-to-Video Model

Text-to-Video (T2V) generation has attracted significant attention for its ability to synthesize realistic videos from textual descriptions. However, existing models struggle to balance computational efficiency and high visual quality, particularly on resource-limited devices, e.g.,iGPUs and mobile phones. Most prior work prioritizes visual fidelity while overlooking the need for smaller, more efficient models suitable for real-world deployment. To address this challenge, we propose a lightweight T2V framework, termed Hummingbird, which prunes existing models and enhances visual quality through visual feedback learning. Our approach reduces the size of the U-Net from 1.4 billion to 0.7 billion parameters, significantly improving efficiency while preserving high-quality video generation. Additionally, we introduce a novel data processing pipeline that leverages Large Language Models (LLMs) and Video Quality Assessment (VQA) models to enhance the quality of both text prompts and video data. To support user-driven training and style customization, we publicly release the full training code, including data processing and model training. Extensive experiments show that our method achieves a 31X speedup compared to state-of-the-art models such as VideoCrafter2, while also attaining the highest overall score on VBench. Moreover, our method supports the generation of videos with up to 26 frames, addressing the limitations of existing U-Net-based methods in long video generation. Notably, the entire training process requires only four GPUs, yet delivers performance competitive with existing leading methods. Hummingbird presents a practical and efficient solution for T2V generation, combining high performance, scalability, and flexibility for real-world applications.

cs.CV↗

arXiv2025

E-MMDiT: Revisiting Multimodal Diffusion Transformer Design for Fast Image Synthesis under Limited Resources

Diffusion models have shown strong capabilities in generating high-quality images from text prompts. However, these models often require large-scale training data and significant computational resources to train, or suffer from heavy structure with high latency. To this end, we propose Efficient Multimodal Diffusion Transformer (E-MMDiT), an efficient and lightweight multimodal diffusion model with only 304M parameters for fast image synthesis requiring low training resources. We provide an easily reproducible baseline with competitive results. Our model for 512px generation, trained with only 25M public data in 1.5 days on a single node of 8 AMD MI300X GPUs, achieves 0.66 on GenEval and easily reaches to 0.72 with some post-training techniques such as GRPO. Our design philosophy centers on token reduction as the computational cost scales significantly with the token count. We adopt a highly compressive visual tokenizer to produce a more compact representation and propose a novel multi-path compression module for further compression of tokens. To enhance our design, we introduce Position Reinforcement, which strengthens positional information to maintain spatial coherence, and Alternating Subregion Attention (ASA), which performs attention within subregions to further reduce computational cost. In addition, we propose AdaLN-affine, an efficient lightweight module for computing modulation parameters in transformer blocks. Our code is available at https://github.com/AMD-AGI/Nitro-E and we hope E-MMDiT serves as a strong and practical baseline for future research and contributes to democratization of generative AI models.

cs.CV↗

arXiv2025

Bohdi: Heterogeneous LLM Fusion with Automatic Data Exploration

Heterogeneous Large Language Model (LLM) fusion integrates the strengths of multiple source LLMs with different architectures into a target LLM with low computational overhead. While promising, existing methods suffer from two major limitations: 1) reliance on real data from limited domain for knowledge fusion, preventing the target LLM from fully acquiring knowledge across diverse domains, and 2) fixed data allocation proportions across domains, failing to dynamically adjust according to the target LLM's varying capabilities across domains, leading to a capability imbalance. To overcome these limitations, we propose Bohdi, a synthetic-data-only heterogeneous LLM fusion framework. Through the organization of knowledge domains into a hierarchical tree structure, Bohdi enables automatic domain exploration and multi-domain data generation through multi-model collaboration, thereby comprehensively extracting knowledge from source LLMs. By formalizing domain expansion and data sampling proportion allocation on the knowledge tree as a Hierarchical Multi-Armed Bandit problem, Bohdi leverages the designed DynaBranches mechanism to adaptively adjust sampling proportions based on the target LLM's performance feedback across domains. Integrated with our proposed Introspection-Rebirth (IR) mechanism, DynaBranches dynamically tracks capability shifts during target LLM's updates via Sliding Window Binomial Likelihood Ratio Testing (SWBLRT), further enhancing its online adaptation capability. Comparative experimental results on a comprehensive suite of benchmarks demonstrate that Bohdi significantly outperforms existing baselines on multiple target LLMs, exhibits higher data efficiency, and virtually eliminates the imbalance in the target LLM's capabilities. Our code is available at https://github.com/gjq100/Bohdi.git.

cs.LG↗

arXiv2025

mLR: Scalable Laminography Reconstruction based on Memoization

ADMM-FFT is an iterative method with high reconstruction accuracy for laminography but suffers from excessive computation time and large memory consumption. We introduce mLR, which employs memoization to replace the time-consuming Fast Fourier Transform (FFT) operations based on an unique observation that similar FFT operations appear in iterations of ADMM-FFT. We introduce a series of techniques to make the application of memoization to ADMM-FFT performance-beneficial and scalable. We also introduce variable offloading to save CPU memory and scale ADMM-FFT across GPUs within and across nodes. Using mLR, we are able to scale ADMM-FFT on an input problem of 2Kx2Kx2K, which is the largest input problem laminography reconstruction has ever worked on with the ADMM-FFT solution on limited memory; mLR brings 52.8% performance improvement on average (up to 65.4%), compared to the original ADMM-FFT.

cs.DC↗

arXiv2025

Development of a 10.8-eV Tabletop Femtosecond Laser with Tunable Polarization for High-Resolution Angle-Resolved Photoemission Spectroscopy

The development of extreme ultraviolet sources is critical for advancing angleresolved photoemission spectroscopy (ARPES), a powerful technique for probing the electronic structure of materials. Here, we report the construction of a tabletop 10.8-eV femtosecond laser through cascaded third-harmonic generation, which operates at a repetition rate of 1 MHz and delivers a photon flux of approximately 1012 photons/s. The system achieves a high energy resolution of approximately 11.8 meV and tunable polarization. This flexibility enables detailed studies of orbitaland (pseudo)spin characteristics in quantum materials. We demonstrate the capabilities of this laser-ARPES system by investigating several prototypical materials, showcasing its potential for elucidating complex phenomena in quantum materials.

cond-mat.mtrl-sci↗

arXiv2025

Factor-guided estimation of large covariance matrix function with conditional functional sparsity

This paper addresses the fundamental task of estimating covariance matrix functions for high-dimensional functional data/functional time series. We consider two functional factor structures encompassing either functional factors with scalar loadings or scalar factors with functional loadings, and postulate functional sparsity on the covariance of idiosyncratic errors after taking out the common unobserved factors. To facilitate estimation, we rely on the spiked matrix model and its functional generalization, and derive some novel asymptotic identifiability results, based on which we develop DIGIT and FPOET estimators under two functional factor models, respectively. Both estimators involve performing associated eigenanalysis to estimate the covariance of common components, followed by adaptive functional thresholding applied to the residual covariance. We also develop functional information criteria for model selection with theoretical guarantees. The convergence rates of involved estimated quantities are respectively established for DIGIT and FPOET estimators. Numerical studies including extensive simulations and a real data application on functional portfolio allocation are conducted to examine the finite-sample performance of the proposed methodology.

stat.ME↗

arXiv2025

Evolving LLMs' Self-Refinement Capability via Synergistic Training-Inference Optimization

Self-Refinement refers to a model's ability to revise its own responses to produce improved outputs. This capability can also serve as a fundamental mechanism for Self-Improvement, for example, by reconstructing datasets with refined results to enhance intrinsic model performance. However, our comprehensive experiments reveal that large language models (LLMs) show no clear evidence of inherent Self-Refinement and may even experience response quality degradation after Self-Refinement. To address this issue, we propose EVOLVE, a simple and effective framework for eliciting and tracking the evolution of Self-Refinement through iterative training. We first explore optimization methods during training to activate the model's Self-Refinement capability. Then, at inference, we investigate various generation strategies to further enhance and utilize Self-Refinement while supplying the necessary data for training. Through synergistic optimization of training and inference stages, we continually evolve the model's Self-Refinement ability, enabling it to better refine its own responses. Moreover, we demonstrate the potential of leveraging Self-Refinement to achieve broader Self-Improvement of intrinsic model abilities. Experiments show that the evolved Self-Refinement ability enables the Llama-3.1-8B base model to surpass GPT-4o, achieving 62.3% length-controlled and 63.3% raw win rates on AlpacaEval 2, and 50.3% on Arena-Hard. It also generalizes effectively to out-of-domain reasoning tasks, improving performance on mathematical reasoning benchmarks such as GSM8K and MATH.

cs.CL↗

arXiv2025

TWC-SLAM: Multi-Agent Cooperative SLAM with Text Semantics and WiFi Features Integration for Similar Indoor Environments

Multi-agent cooperative SLAM often encounters challenges in similar indoor environments characterized by repetitive structures, such as corridors and rooms. These challenges can lead to significant inaccuracies in shared location identification when employing point cloud-based techniques. To mitigate these issues, we introduce TWC-SLAM, a multi-agent cooperative SLAM framework that integrates text semantics and WiFi signal features to enhance location identification and loop closure detection. TWC-SLAM comprises a single-agent front-end odometry module based on FAST-LIO2, a location identification and loop closure detection module that leverages text semantics and WiFi features, and a global mapping module. The agents are equipped with sensors capable of capturing textual information and detecting WiFi signals. By correlating these data sources, TWC-SLAM establishes a common location, facilitating point cloud alignment across different agents' maps. Furthermore, the system employs loop closure detection and optimization modules to achieve global optimization and cohesive mapping. We evaluated our approach using an indoor dataset featuring similar corridors, rooms, and text signs. The results demonstrate that TWC-SLAM significantly improves the performance of cooperative SLAM systems in complex environments with repetitive architectural features.

cs.RO↗

arXiv2025

STG-Avatar: Animatable Human Avatars via Spacetime Gaussian

Realistic animatable human avatars from monocular videos are crucial for advancing human-robot interaction and enhancing immersive virtual experiences. While recent research on 3DGS-based human avatars has made progress, it still struggles with accurately representing detailed features of non-rigid objects (e.g., clothing deformations) and dynamic regions (e.g., rapidly moving limbs). To address these challenges, we present STG-Avatar, a 3DGS-based framework for high-fidelity animatable human avatar reconstruction. Specifically, our framework introduces a rigid-nonrigid coupled deformation framework that synergistically integrates Spacetime Gaussians (STG) with linear blend skinning (LBS). In this hybrid design, LBS enables real-time skeletal control by driving global pose transformations, while STG complements it through spacetime adaptive optimization of 3D Gaussians. Furthermore, we employ optical flow to identify high-dynamic regions and guide the adaptive densification of 3D Gaussians in these regions. Experimental results demonstrate that our method consistently outperforms state-of-the-art baselines in both reconstruction quality and operational efficiency, achieving superior quantitative metrics while retaining real-time rendering capabilities. Our code is available at https://github.com/jiangguangan/STG-Avatar

cs.CV↗

arXiv2025

Latent Harmony: Synergistic Unified UHD Image Restoration via Latent Space Regularization and Controllable Refinement

Ultra-High Definition (UHD) image restoration faces a trade-off between computational efficiency and high-frequency detail retention. While Variational Autoencoders (VAEs) improve efficiency via latent-space processing, their Gaussian constraint often discards degradation-specific high-frequency information, hurting reconstruction fidelity. To overcome this, we propose Latent Harmony, a two-stage framework that redefines VAEs for UHD restoration by jointly regularizing the latent space and enforcing high-frequency-aware reconstruction.In Stage One, we introduce LH-VAE, which enhances semantic robustness through visual semantic constraints and progressive degradation perturbations, while latent equivariance strengthens high-frequency reconstruction.Stage Two jointly trains this refined VAE with a restoration model using High-Frequency Low-Rank Adaptation (HF-LoRA): an encoder LoRA guided by a fidelity-oriented high-frequency alignment loss to recover authentic details, and a decoder LoRA driven by a perception-oriented loss to synthesize realistic textures. Both LoRA modules are trained via alternating optimization with selective gradient propagation to preserve the pretrained latent structure.At inference, a tunable parameter α enables flexible fidelity-perception trade-offs.Experiments show Latent Harmony achieves state-of-the-art performance across UHD and standard-resolution tasks, effectively balancing efficiency, perceptual quality, and reconstruction accuracy.

cs.CV↗

arXiv2025

Týr-the-Pruner: Structural Pruning LLMs via Global Sparsity Distribution Optimization

Structural pruning enhances hardware-agnostic inference efficiency for large language models (LLMs) yet often fails to maintain comparable performance. Local pruning performs efficient layer-by-layer compression but ignores global topology. Although global pruning aims to identify an optimal sparse model, intuitive methods typically adopt a two-stage paradigm that first evaluates substructure saliency and then applies global pruning, which ignores inter-structure dependencies and fails to achieve end-to-end optimization. To address these limitations, we propose Týr-the-Pruner, an efficient end-to-end search-based global structural pruning framework. This framework constructs a supernet by repeatedly applying local pruning across a range of sparsity ratios to each layer in an LLM, with the core goal of determining the optimal sparsity distribution under a target overall sparsity ratio. Concretely, we introduce an effective local pruning and an expectation error accumulation approach to improve supernet construction. Furthermore, we employ an iterative prune-and-search strategy with coarse-to-fine sparsity granularity to ensure efficient search convergence. Experimental results show that Týr-the-Pruner achieves state-of-the-art structural pruning, retaining 97% of the dense model's performance while removing a challenging 50% of Llama-3.1-70B's parameters. Code will be available at https://github.com/AMD-AGI/Tyr-the-Pruner.

cs.LG↗

arXiv2025

SolverLLM: Leveraging Test-Time Scaling for Optimization Problem via LLM-Guided Search

Large Language Models (LLMs) offer promising capabilities for tackling complex reasoning tasks, including optimization problems. However, existing methods either rely on prompt engineering, which leads to poor generalization across problem types, or require costly supervised training. We introduce SolverLLM, a training-free framework that leverages test-time scaling to solve diverse optimization problems. Rather than solving directly, SolverLLM generates mathematical formulations and translates them into solver-ready code, guided by a novel Monte Carlo Tree Search (MCTS) strategy. To enhance the search process, we modify classical MCTS with (1) dynamic expansion for adaptive formulation generation, (2) prompt backpropagation to guide exploration via outcome-driven feedback, and (3) uncertainty backpropagation to incorporate reward reliability into decision-making. Experiments on six standard benchmark datasets demonstrate that SolverLLM outperforms both prompt-based and learning-based baselines, achieving strong generalization without additional training.

cs.LG↗

arXiv2025

Reconnection-driven Decaying Pulsations Modulated by Slow Magnetoacoustic Waves

Decaying pulsations have been simultaneously detected in the low-energy X-rays of solar/stellar flares, which are supposed to be associated with standing slow magnetoacoustic or kink-mode waves. The physical mechanism behind rapidly decaying remains unknown. We present the detection of quasi-periodic pulsations (QPPs) with rapidly decaying in high-energy emissions produced in two major flares on 10 January and 14 May 2024. Using empirical mode decomposition, decaying QPPs are identified in hard X-ray and microwave emissions during the flare impulsive phase, suggesting a process of oscillatory magnetic reconnection. The quasi-periods and decay times are determined by a damped harmonic function, which are approximately 177$\pm$8 s (249$\pm$25 s) and 118$\pm$4 s (124$\pm$5 s), respectively. The restructured X-ray images reveal double footpoints connected by hot flare loops. Their phase speeds are estimated to about 400 km/s and 670 km/s, both below the local sound speed in high-temperature plasmas, indicating the presence of slow-mode waves in hot flare loops. We perform coronal diagnostics based on standing slow-mode waves and derive key physical parameters, including the polytropic index, the thermal ratio, viscous ratio and radiation ratio, which are consistent with previous results. Our observations support that the decaying QPPs are triggered by oscillatory magnetic reconnection that is modulated by standing slow magnetoacoustic waves, with their rapid decay attributable to a co-effect of viscous damping and localized magnetic reconnection rate.

astro-ph.SR↗

arXiv2025

A Survey of Reinforcement Learning for Large Reasoning Models

In this paper, we survey recent advances in Reinforcement Learning (RL) for reasoning with Large Language Models (LLMs). RL has achieved remarkable success in advancing the frontier of LLM capabilities, particularly in addressing complex logical tasks such as mathematics and coding. As a result, RL has emerged as a foundational methodology for transforming LLMs into LRMs. With the rapid progress of the field, further scaling of RL for LRMs now faces foundational challenges not only in computational resources but also in algorithm design, training data, and infrastructure. To this end, it is timely to revisit the development of this domain, reassess its trajectory, and explore strategies to enhance the scalability of RL toward Artificial SuperIntelligence (ASI). In particular, we examine research applying RL to LLMs and LRMs for reasoning abilities, especially since the release of DeepSeek-R1, including foundational components, core problems, training resources, and downstream applications, to identify future opportunities and directions for this rapidly evolving area. We hope this review will promote future research on RL for broader reasoning models. Github: https://github.com/TsinghuaC3I/Awesome-RL-for-LRMs

cs.CL↗